water research 43 (2009) 2717–2727

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

GISCOD: General Integrated Solid Waste Co-Digestion model

Usama Zahera,*, Rongping Lia,b, Ulf Jeppssonc, Jean-Philippe Steyerd, Shulin Chena
a
Department of Biological Systems Engineering, Washington State University, P.O. Box 646120, Pullman, WA 99164-6120, USA
b
Department of Environmental Engineering, Beijing University of Chemical Technology, 100029 Beijing, PR China
c
Department of Industrial Electrical Engineering and Automation, Lund University, Box 118, SE-22100, Lund, Sweden
d
INRA, UR 50, Laboratoire de Biotechnologie de l’Environnement, Avenue des Etangs, F-11100 Narbonne, France

article info abstract

Article history: This paper views waste as a resource and anaerobic digestion (AD) as an established
Received 22 December 2008 biological process for waste treatment, methane production and energy generation.
Received in revised form A powerful simulation tool was developed for the optimization and the assessment of
11 March 2009 co-digestion of any combination of solid waste streams. Optimization was aimed to
Accepted 14 March 2009 determine the optimal ratio between different waste streams and hydraulic retention time
Published online 21 March 2009 by changing the digester feed rates to maximize the biogas production rate. Different
model nodes based on the ADM1 were integrated and implemented on the Matlab-Simu-
Keywords: linkÒ simulation platform. Transformer model nodes were developed to generate detailed
ADM1 input for ADM1, estimating the particulate waste fractions of carbohydrates, proteins,
Co-digestion lipids and inerts. Hydrolysis nodes were modeled separately for each waste stream. The
Hydrolysis fluxes from the hydrolysis nodes were combined and generated a detailed input vector to
Integrated modeling the ADM1. The integrated model was applied to a co-digestion case study of diluted dairy
Solid waste manure and kitchen wastes. The integrated model demonstrated reliable results in terms
Transformer model of calibration and optimization of this case study. The hydrolysis kinetics were calibrated
for each waste fraction, and led to accurate simulation results of the process and prediction
of the biogas production. The optimization simulated 200,000 days of virtual experimental
time in 8 h and determined the feedstock ratio and retention time to set the digester
operation for maximum biogas production rate.
Published by Elsevier Ltd.

1. Introduction (Garcı́a de Cortázar and Monzón, 2007), and suggested as

This paper presents GISCOD, a general integrated solid waste
co-digestion model. The main goal of this study was to
develop and test a simulation tool of the anaerobic digestion
(AD) process that is applicable to any combinations of waste
streams using the simulation platform Matlab-SimulinkÒ. The
MatlabÒ simulation platform was used for implementation of
the risk assessment of gas emissions from solid waste incin-
erators (Kumar et al., 2009) and modeling solid waste landfills
a common interface model for solid waste management (bou
Najm and El-Fadel, 2004).
A general co-digestion model is needed to support opera-
tion decisions at full-scale plants and to assist co-digestion
research. The AD process is a widely applicable technology to
treat and convert an organic waste stream to methane for
green energy production. At wastewater treatment plants,
trucked-in wastes are digested with wastewater sludge for
renewable energy production (Wallis et al., 2008; Zupancic

* Corresponding author. Fax: þ1 509 335 2722.

E-mail addresses: zaheru@wsu.edu (U. Zaher), ulf.jeppsson@iea.lth.se (U. Jeppsson), steyer@supagro.inra.fr (J.-P. Steyer), chens@
wsu.edu (S. Chen).
0043-1354/$ – see front matter Published by Elsevier Ltd.
doi:10.1016/j.watres.2009.03.018
2718 water research 43 (2009) 2717–2727
et al., 2008) as part of municipal policies for climate change
mitigation and reduction of green house gas emissions. Biogas
plants co-digest different solid waste feedstock to increase
biogas production. However, random or heuristic decision on
the ratio between waste streams or feedstock to full-scale
plants often lead to process upset and significant reduction of
methane production (Steyer et al., 2006). The general model
would support such full-scale operation decisions. Significant
research effort was devoted during the last 5 years to study the
co-digestion of different combinations of municipal, indus-
trial, agricultural and farming waste streams. A general model
is needed to define optimal co-digestion experiments sparing
research efforts of experimental trials, and to simulate AD
improvement mechanisms that are achieved by co-digestion
such as buffered pH, reduced inhibition, improved hydrolysis
and/or adjusted C/N ratio. Improvement mechanisms of
co-digestion can be simulated by the ADM1, International
Water Association Anaerobic Digestion Model number 1,
which was developed by the task group on anaerobic digestion
(Batstone et al., 2002). However, the ADM1 application has
practical problems related to the characterization of the
digester feedstock and the associated model definition of the
enzymatic disintegration and hydrolysis steps.
A generalized and separate approach is required to solve
the solid waste characterization problems compared to Acti-
vated Sludge (AS) for two reasons. Firstly, ADM1 is considering
constant composition of particulates with fixed fraction
parameters to carbohydrates, proteins, lipids and inerts. On
the contrary, solid wastes are heterogenic and dynamically
changing in composition. Secondly, the lumped composite
particulate model component is used as the first model input
and, simultaneously, as a product from the model decay
processes. This implies that the fraction parameters and
hydrolysis rates of the feed substrate should match the
composition and hydrolysis rates of the decaying biomass. In
fact, the ADM1 was originally developed with focus on the
application of AS digestion assuming similar composition of
the aerobic and anaerobic bacteria. Under this assumption,
there is no conflict between the feed substrate and the
produced substrate from decaying bacteria. In this particular
case, cell lysis (disintegration) is the limiting hydrolysis step.
Such an assumption was proven to be consistent for plant-
wide modeling since the AS inert fraction remains inert under
anaerobic conditions (Ekama et al., 2007).
In previous applications of the ADM1, fraction parameters
were estimated from experimental data (Fezzani and Cheikh,
2008a,b) or evaluated as function of VS influx (Lübken et al.,
2007). Using a priori expert knowledge about expected waste-
water characteristics and experimental measurements to
estimate fraction parameters is generally applied for
modeling wastewater treatment systems (Grau et al., 2007).
Applying such a procedure to co-digestion is not feasible since
it is difficult to find unique parameter values that are appli-
cable to all possible combinations and ratios of solid wastes
together with decaying anaerobic biomass.
Parameter estimation problems and use of fraction
parameters could be avoided using a dynamic interface to
ADM1 to simulate AD of animal manure and solid waste
(Zaher and Chen, 2006). The interface procedure was validated
by comparing the estimated carbohydrates, proteins, lipids
and inerts concentrations with the proximate analysis of 17
solid wastes (Zaher et al., 2009). In the research work pre-
sented in this paper, the interface procedure is generalized
and implemented with GISCOD in Matlab-Simulink as
a general transformer model that interface ADM1 to any
combination of co-digested wastes. The influxes of the model
components from each waste are evaluated dynamically. The
hydrolysis parameters are considered separately for each
waste and uncoupled from the hydrolysis of the decaying
biomass. Therefore, the GISCOD modeling tool is generalized
to study the co-digestion of any combination of different
wastes and to evaluate their independent hydrolysis rates and
operation settings, i.e. their optimal feed ratio and hydraulic
retention time (HRT).
2. Methods
2.1. Process model
The AD process was modeled using the ADM1 (Batstone et al.,
2002) as a basis with phased implementation to separate the
enzymatic hydrolysis of solid wastes from the metabolic
reactions utilizing soluble substrates. The ADM1 model starts
with a disintegration step of composite particulate material,
i.e. decomposition of feed or decaying biosolids according to
their predefined fractions and composition of carbohydrates,
proteins, fat (lipids) and inerts. The second step is enzymatic
hydrolysis of disintegrated carbohydrates, proteins and fat
(lipids), which is the start of the corresponding three path-
ways of anaerobic degradation. The anaerobic degradation is
done in three main stepsdacidogenesis, acetogenesis and
methanogenesis. The degradation steps are modeled by
uptake kinetics of different substrates by seven bacterial
groups. The decay processes of the seven bacterial groups are
also considered and the decaying particulates are sent back to
the disintegration step.
The implemented GISCOD model shown in Fig. 1 is
generalized to consider the degradation of any other wastes
that are different in composition compared to the assumed
biosolids (i.e. decaying bacteria). Each waste would have
different fractions of carbohydrates, proteins, lipids and inerts
that may be changing dynamically (Lübken et al., 2007). Each
waste would also have different hydrolysis rates of carbohy-
drates, proteins and lipids (Fezzani and Cheikh, 2008a,b).
Carbohydrates, proteins and lipids hydrolysis of each waste is
considered in separate model nodes. The disintegration step
was not considered for solid wastes assuming that enzymes
can diffuse in the woven structure of wastes and hydrolysis
would take place before disintegration. No cell lysis is required
for solid wastes compared to AS or decaying bacteria. The
hydrolysis products are combined and used as input to
a single digestion node where all biological reactions of ADM1
are activated. The non-hydrolyzed fractions are fed through
the digestion node as a dummy vector and hydrolysis kinetics
in the digestion node are only applied to the decaying
biosolids. The complete structure of ADM1 is considered in the
hydrolysis nodes to allow future expansion of the co-digestion
model considering more complex hydrolysis kinetics. The
other biological reactions are deactivated for the hydrolysis
 water research 43 (2009) 2717–2727
Fig. 1 – The GISCOD model in Matlab-Simulink.
nodes simply by assuming zero uptake, disintegration, decay
and gas transfer rates.
In addition to the biological reactions, the ADM1 imple-
mentation considers the chemical equilibrium of all ions to
evaluate the pH change. The chemical equilibrium of volatile
fatty acids (VFA), the carbon and nitrogen systems is solved
externally once for all hydrolysis and digestion nodes. The
solution of chemical equilibrium is performed algebraically
according the ADM1–DAE implementation (Rosen et al., 2006).
The ADM1–DAE implementation removes stiffness from the
original ADM1 Ordinary Differential Equations (ODE) system
to simulate rapid dynamic changes in the anaerobic digestion
process, e.g. due to changing composition of the digester
feedstock.
2.2. Transformer model
A general transformer model to interface ADM1 to different
solid waste streams was programmed in C and incorporated
in the GISCOD Matlab-Simulink model as a C-MEX S-Function.
The general transformer model is based on the ADM1 inter-
face to solid wastes (Zaher and Chen, 2006; Zaher et al., 2009).
The transformer model combines the advantages of previous
interfacing methodologies applied to ADM1.
2.2.1. Implemented interfacing advantages
The general transformer model represents an enhancement
of the Continuity Based Interfacing Methodology (CBIM)
(Vanrolleghem et al., 2005). The CBIM applies Chemical
Oxygen Demand (COD) balance, charge balance and elemental
continuity to all macronutrient elements CHNOP to connect
different models (Volcke et al., 2006; Zaher et al., 2007). The
CBIM in the general transformer is applied to interface the
ADM1 to practical characteristics of solid wastes.
2719
Kleerebezem and van Loosdrecht (2006) used practical char-
acteristics such as COD, Total Kjeldahl Nitrogen (TKN), etc. to
characterize the ADM1 influent. They assumed the digester’s
feedstock as a single composite particulate (Xc) with constant
composition and used the practical characteristics to estimate
ADM1 fraction parameters that distribute Xc after disinte-
gration to particulate components of carbohydrates, proteins
and lipids. The use of the fraction parameters does not allow
dynamic simulation due to changes in the feedstock compo-
sition. The transformer model applies CBIM to estimate the
influxes to ADM1 and avoids the overuse of fraction parame-
ters to allow dynamic simulation. The transformer model
robustness is increased by updating the CBIM procedure to
maximize the conversions to ADM1 components in a pre-
defined order. COD and charge balances, and the continuity of
all CHNOP elements are checked after the conversion of each
component. Such an ordered maximization procedure was
suggested by Copp et al. (2003) to interface Activated Sludge
Model no.1 (Henze et al., 1987) ASM1 with ADM1, maintaining
the COD and N balances. Most recently, Nopens et al. (in press)
modified the Copp et al. (2003) ASM1–ADM1 interface. They
increased the robustness of the ASM1–ADM1 conversions by
changing the maximization order of ADM1 components for
the co-digestion of secondary sludge (from ASM1) with
primary sludge (from primary settler). The conversions from
ASM1 to ADM1 were extended to include carbohydrates,
proteins and lipids instead of Xc. Proteins were maximized
using Total Kjeldahl Nitrogen (TKN) and the remaining
particulate COD was distributed between carbohydrate and
lipids using fraction parameters. Extending the balance rela-
tions in the transformer model eliminates the use of fraction
parameters. The transformer model is upgraded in the
implementation with GISCOD to allow the user to change the
maximization order of ADM1 components without changing
2720 water research 43 (2009) 2717–2727
the transformer model algorithm. The maximization order is
implemented as a one-dimensional array parameter to the
transformer S-function to increase the generality of the
transformer model for the application to any solid waste.
2.2.2. Transformer algorithm
The transformer model transforms a set of practical
measurements to the input vector of ADM1 according to the
stoichiometry presented in Table 1. Table 1 consists of four
panes. The lower two panes shows the assumed composition
of the practical measurements on the left, components 1 to 11,
and the composition of the estimated ADM1 input compo-
nents on the right. The upper two panes represent the stoi-
chiometry nj;k for the conversions j and the elements k. The
stoichiometry is evaluated by mass and charge balances
according to Eq. (1). The stoichiometry matrix is uploaded to
the Matlab work space as a two-dimensional array parameter
to transformer model S-function.
X X
nj;k ij;Comp ¼ 0 with Comp ¼ Thod; C; N; H; O; e
k 1
The transformation step in CBIM was changed to include the
ordered maximization procedure. The original transformation
of CBIM is generated by Eqs. (2) and (3). A set of algebraic
equations is generated by Eq. (2) to map the influxes to vector
rj, j ¼ 1:n where n is the number of conversions, using the
stoichiometry in the left pane of the transformation matrix,
i.e., for k ¼ 1:P where P is the number of practical measure-
ments. Then Eq. (3) calculates the outfluxes from rj using the
stoichiometry in the right pane of the transformation matrix,
i.e., k ¼ P þ 1:P þ Q where Q is the number of the estimated
composition components.
X
n
nj;k rj ¼ Influxk for k ¼ 1 : P
j¼1
X
n solid wastes as well as all model parameters are arranged in
Outfluxk ¼ nj;k rj for k ¼ P þ 1 : P þ Q
j¼1
In the implementation for the co-digestion model Eq. (2) is
replaced by a maximization procedure according to Zaher
et al. (2009) to increase the transformer robustness, to conceal
(correct) possible errors in the practical measurements and to
maintain the elemental mass balance during the conversions.
The elements of the vector rj are maximized in a predefined
order to make sure that the elemental influxes sourced by the
input of practical measurements are sufficient before calcu-
lating the next element of rj. A predefined order of rz, z ¼ 1:10,
which corresponds to j ¼ (10, 5:9, 4, 3, 1, 2), maximizes the
conversion to inert particulates, volatile fatty acids, sugars,
lipids, proteins, carbohydrates, and then inorganic compo-
nents. This maximization order is uploaded before simulation
to the Matlab work space as a parameter to the transformer
model S-function. Thus, the maximization order can be easily
changed by the user. The maximization is done according to
the following steps:
1. rz is calculated using Eq. (4) as a function of the influx of the
most correlated measurement k, i.e., corresponding to the
unique value of nz,k ¼ -1 at each conversion;
2. rz is verified using the conditions imposed by Eq. (5). If
shown true, the next rzþ1 was calculated starting from step
1 above;
3. If shown false, rz is changed and calculated according to Eq.
(6), the rz calculation is then terminated and other rates (ri,
i ¼ z þ 1:n) are assigned a value of 0;
4. Any remaining fluxes are added to the relevant inorganic
components; and accordingly,
5. All practical measurements are mapped to the new vector
r. The output flux of substrate composition is then calcu-
lated using Eq. (3).
P !
Influxk  z1 i¼1 ni;k ri
rz ¼ (4)
nz;k
z
(1) nz;k rz < Influxk for k ¼ 1 : P (5)
 
 
 P 
Influxk  z1 n r 

rz ¼ min i¼1 i;k i 
for k ¼ 1 : P (6)
n 
 z;k 
 
2.3. Integrated co-digestion model
The different models integrated in GISCOD are written in C
and compiled in Matlab as MEX S-functions to run simulations
and optimizations using the Matlab-Simulink platform and its
toolboxes. The compiled version of the model works with
most Matlab-Simulink (release 14) installations on Windows
(2)
XP and VISTA operating systems.
The practical characteristics and flows of all different
(3) Microsoft Excel file. All inputs, initial states and parameters
to the co-digestion models are read from the Excel file into
the Matlab work space using an automated Matlab script.
The simulation starts from Simulink after configuring the
numerical solution using any variable step solver that is
available in Simulink. Fig. 1 shows the scheme of GISCOD in
Matlab-Simulink. Practical characteristics and flows of each
solid waste are inputs from the workspace to the transformer
model nodes. The practical characteristics are converted to
the complex composition of the ADM1 input state vector and
assigned to the input of separate hydrolysis nodes. The
hydrolysis output signals are rearranged by the combiner
model, which generates the input to the ADM1 node. The
combiner model divides the solid wastes AD process into an
enzymatic hydrolysis phase in the hydrolysis nodes only and
an uptake phase of the hydrolysis products in the ADM1
node. Thus, Solids Residence Time (SRT) of each waste is
considered separately for each hydrolysis node according to
the time its particulate components are allowed to stay in the
digester (i.e. according mixing patterns) in addition to the
time of any pre-hydrolysis steps. The combiner node passes
the non-hydrolyzed particulates as dummy variables to the
ADM1 and sums other variables on the basis of fluxes from
Table 1 – Calculated transformation and composition matrices of the transformer model.

water research 43 (2009) 2717–2727

2721
2722 water research 43 (2009) 2717–2727

both waste streams. In the ADM1 node, non-hydrolyzed
portions are not subject again to the hydrolysis kinetics and
the hydrolysis in the ADM1 node is only considered for
particulate fractions of the decaying biosolids (bacteria).
Thus, the digester out-flux contains non-hydrolyzed carbo-
hydrates, proteins and lipids originating from the solid
wastes in addition to the corresponding components result-
ing from decaying biosolids. Thus the mass balance is
maintained. The ADM1 is solved at each time step and the
output is stored in the Matlab workspace. The chemical
equilibrium is solved in the ADM1 model node and the
evaluated ions and pH are shared with the hydrolysis nodes.
The pH calculation is linked to the hydrolysis nodes to allow
future extension of the hydrolysis kinetics to reflect the pH
dependency of the hydrolysis. Although the optimal pH of
methanogenesis is around pH 7.0, the optimum pH of
hydrolysis and acidogenesis is between pH 5.5 and 6.5 (Ward
et al., 2008).
2.4. Calibration and optimization case study
GISCOD robustness and simulation speed were tested by
running the model in parameter estimation and optimization
algorithms. Parameter estimation was done using SimulinkÒ
Parameter EstimationÔ software and the simplex optimiza-
tion algorithm (Nelder and Mead, 1965). Two experiments of
digesting manure alone and manure with kitchen waste were
performed to calibrate the hydrolysis parameters for each
waste. Both waste average characteristics are listed in Table 2.
Only the indicated 11 characteristics are needed as model
inputs. It was not possible to digest food waste alone due to
acidification and pH drop. Both wastes were homogenized
and kept frozen in batches that were only thawed before
feeding. The only degree of freedom used during the experi-
ment was the daily feed rates, which were varied for each
experiment according to the profiles shown in Fig. 2. The
reactors for both experiments were completely mixed and
arranged to have a hydrolysis step of 0.6 L volume followed by
a digestion step of 2 L. All reactors were kept at 35  C. The gas
production from both steps was used for calibration. First, the
manure hydrolysis parameters were estimated from the
manure only digestion experiment. Secondly, the kitchen
waste hydrolysis parameters were estimated from the co-
digestion experiment. Carbohydrates, proteins and lipids
were analyzed for each waste to validate the transformer
predictions. Carbohydrates were quantified by sequential
extraction using neutral and acid detergent, followed by
strong acid extraction. Proteins were analyzed by the Lowry
colorimetric method calibrated on bovine serum albumin.
The lipids content was determined by a Soxhlet method using
petroleum ether for extraction.
Optimization of the solid waste ratio and HRT was done by
simulating several virtual experiments using the calibrated
model. The optimal ratio and HRT were determined by
comparing the steady state biogas flow rate from such virtual
experiments. Virtual experiments of 200 cases were simu-
lated varying the ratio of kitchen waste, flow and methano-
genic reactor volume. Ten retention times were considered 5,
7.5, 10, 15, 20, 50, 75, 100, 150 and 200 days. The kitchen waste
ratio was varied from 5% to 100% in 5% increments. The
hydrolysis volume was 2 L for all the simulated cases. Two
methanogenic volumes were considered: 2 L with HRT
20 days and 20 L for longer HRT. Each case was simulated
until the gas flow rate reached a steady state after 1000 days
of simulation time, i.e. a total virtual experimental time of
200,000 days.

Table 2 – Characteristics of diluted manure and kitchen waste.

Characteristics Co-digestion Unit Diluted manure waste Kitchen waste
model input no.

Total Chemical Oxygen Demand (CODt) (gCOD m3) 27217 380647

Particulate COD (CODp) 1 (gCOD m3) 23550 368400
Soluble COD (CODs) (gCOD m3) 3667 12247
Soluble COD without VFA COD(CODs-VFA) 2 (gCOD m3) 2521 3500
Volatile Fatty Acids (VFA) 3 (gCOD m3) 1146 8747
Total Carbon (TC) (gC m3) 10064 139760
Total Organic Carbon (TOC) 4 (gC m3) 9340 139280
Total Inorganic Carbon (TIC) 9 (mol HCO 3
3 m ) 60 40
Total Kheldal Nitrogen (TKN) (gN m3) 882 15300
Total Organic Nitrogen (Norg) 5 (gN m3) 598 14000
Total Ammonia Nitrogen (TAN) 6 (gN m3) 284 1300
Total Phosphorous (TP) (gP m3) 219 1606
Organic Phosphorus (TP-orthoP) 7 (gP m3) 187 720
Ortho-Phosphate (orthoP) 8 (gP m3) 32 886
Total alkalinity (S cations) 10 (equ m3) 60 25
Total Solids (TS) (g m3) 20697 291000
Fixed Solids (FS) 11 (g m3) 5397 31000
Total Volatile Solids (TVS) (g m3) 15300 260000
Carbohydrate (g m3) 10924  428 153400  11180
Protein (g m3) 4069  367 85800  8320
Lipids (g m3) 306  61.2 20800  2860
 water research 43 (2009) 2717–2727 2723

Manure Flow Experiment 1 Manure Flow Experiment 2

0.7 0.060
Kitchen Waste Flow Experiment 2

0.6
0.050

Kitchen Waste Feed L/day

0.5
Manure flow L/day 0.040

0.4
0.030
0.3

0.020
0.2

0.010
0.1

0 0.000
0 20 40 60 80 100
Time days

Fig. 2 – Daily feed rates implemented in the calibration experiments.

extraction method is an accurate standard method to break

3. Results and discussion
3.1. Transformer output
Among other ADM1 input state variables, carbohydrates,
proteins and lipids were estimated in COD units by the trans-
former model. The ADM1 model uses COD units for organic
components and bacterial species to maintain the COD
balance. The corresponding g/L concentration was evaluated
according to the defined composition of ADM1 components in
Table 1 and compared to the measured concentrations in Fig. 3.
Generally, the estimated and measured concentrations were
consistent in terms of distribution among the three main
particulate components. However, some differences could be
observed for each individual component when comparing the
results for the manure and kitchen waste.
Carbohydrates: Estimated carbohydrates content was
consistent with measured data in the case of manure but it
was higher in the case of kitchen waste. The detergent
the crystal structure of fiber, which is the main form of
carbohydrates in manure. The starch content is high in
kitchen waste but would not be quantified as accurately as
fiber with the same extraction method. Using the carbohy-
drate measurements as a direct input to the ADM1 model
would have introduced an error to the carbon balance kept
within the model. Therefore, using the transformer model
was necessary to keep the carbon balance.
Proteins: Measured and estimated protein contents were
more consistent in the case of kitchen waste as compared to
the case of manure. The measuring method was calibrated
using bovine serum albumin, which is more relevant to the
kind of proteins that normally exist in kitchen wastes, such
as beef or whey. Using the protein measurements for
manure as a direct input to ADM1 model would have
introduced errors to the nitrogen balance. Nitrogen in solids
wastes is mainly sourced by the particulate proteins. The
use of the transformer model maintained the nitrogen
balance.

Fig. 3 – Comparison of the measured and the estimated compositions by the transformer model for diluted manure (left) and
kitchen waste (right).
2724 water research 43 (2009) 2717–2727

Lipids: Lipids were the smallest fraction of particulates in
both wastes. The estimated and measured lipids contents
were relatively inconsistent. On one hand, the estimated
lipids composition was assumed to be in the form of phos-
pholipids but other forms may exist in both wastes. On the
other hand, Soxhlet extraction is highly biased if the sample
matrix is mainly non-lipids (Manirakiza et al., 2001).
Generally, the use of the transformer model within GISCOD
maintains the continuity of COD and elemental mass that are
essential to guarantee accurate and reliable simulation. Direct
measurements of the waste particulate fractions would not
achieve the same reliability. The analytical methods are
dependent on the types of the particulate fractions, which are
unknown for wastes and are often different from the types
defined in the model stoichiometry.
Maintaining accurate carbon and nitrogen balances during
the simulation is necessary since the C:N ratio is a key factor
affecting the co-digestion of different waste streams
(Hartmann and Ahring, 2005; Yen and Brune, 2007; Zhang
et al., 2008; Shanmugam and Horan, 2009). Also, the C and N
elemental continuity preserved in the GISCOD model is
important when linking the AD model to other existing
models of subsequent unit processes or for integrated
assessment. For instance elemental continuity is the key
mechanism to evaluate pH and chemical equilibrium
variables, such as CO2/HCO þ
3 and NH4 /NH3 in the AD process
out-flux. The evaluation of CO2 and NH3 emissions allows
further assessment of subsequent unit processes, such as
emission studies from composting (Paillat et al., 2005; Komilis
and Ham, 2006), drying (Deng et al., 2009) and landfill facilities
(He et al., 2006). Furthermore, estimation of the pH and NHþ 4 as
well as phosphorusdevaluated from the mass balance in the
transformer modeldallows more integrated assessment,
such as studies evaluating added fertility to soils from waste
application (Alvarenga et al., 2007; Kang et al., 2008) or eval-
uating leachate pollution to water bodies (Singh et al., 2005).
3.2. Simulation speed
The simulation speed was kept low despite the added model
complexity of separate hydrolysis and transformation model
nodes. The 73 days simulation of manure digestion required
less than 1 min CPU-time using the ode15s solver in Simulink
with 1E-7 and 1E-6 relative and absolute tolerance,
respectively. The longer and more dynamic experiment no. 2
of co-digestion required 1 min 30 s CPU-time using the same
simulation settings using a standard 3 GHz PC. The efficient
simulation time was achieved because of the separate alge-
braic solution of the chemical equilibrium. The maintenance
of COD and elemental mass balances using the transformer
contributes to the simulation accuracy, which also contrib-
utes to improved simulation speed.
3.3. Calibration of hydrolysis kinetics
With the reliable simulation speed, it was possible to run the
model using the simplex optimization algorithm for calibra-
tion purposes. Fig. 4 shows the predictions of biogas flow rate
after model calibration, which are comparable to the
measurements. The hydrolysis rates of carbohydrates,
proteins and lipids were estimated by fitting the biogas
measurements from experiment 1, digesting diluted manure
only. The estimated rates were 0.019, 0.025, 0.022 d1,
respectively, for diluted manure waste. These rates are
considerably lower compared to the default values of ADM1
(10 d1 for each particulate component) that were originally
designated for the hydrolysis of activated aerobic sludge and
are still used in GISCOD for the hydrolysis of the decaying
anaerobic bacteria after a disintegration step (rate kdis ¼ 0.5). It
is noteworthy that the default hydrolysis rates presented in
the ADM1 in 2002 is now considered to be at least a factor of
ten too large also by the ADM1 Task Group (Batstone, 2008:
personal communication). The low rates indicate that the
digestion of the manure waste was limited by hydrolysis and
that the amount of methane produced was mainly from
soluble COD digestion. When the diluted manure was co-
digested with kitchen waste the biogas production was
significantly increased even at periods of similar HRT in both
experiments, i.e. day 0 to 38 and day 63 to 73. The higher
biogas production was not only due to the higher COD load of
added kitchen waste but also because the particulate fractions
of the kitchen waste were easily hydrolysable. The estimated
hydrolysis rates of the kitchen waste from the second exper-
iment were 5.22, 1.86 and 1.24 d1 for carbohydrates, proteins
and lipids, respectively. This indicates the necessity of sepa-
rating the hydrolysis of both wastes. In a similar co-digestion
study of a fixed ratio 80:20 manure liquids to cow fodder
(Lübken et al., 2007), the best ADM1 simulation of biogas

Co-digestion (simulation) Co-Digestion (data) Manure Only (simulation) Manure Only (data)

8 7.4
Gas flow rate (L/day)

7 7.3
6 7.2
5 7.1
7
pH

4
6.9
3
6.8
2 6.7
1 6.6
0 6.5
0 20 40 60 80 100 120 0 20 40 60 80 100 120
days days

Fig. 4 – Comparison of simulated and measured biogas production (left) and pH (right) after calibration of the hydrolysis
parameters.
 water research 43 (2009) 2717–2727
prediction matched the experimental data at 0.3 d1 hydro-
lysis rate for the three particulate fractions. The slightly
higher hydrolysis rate compared to digesting manure is due to
the addition of the cow fodder. Cow fodder hydrolysis rate is
higher compared to manure that has already been passed
through hydrolysis and digestion in the rumen. For reliable
simulation and prediction of the biogas production at variable
ratios of co-digested wastes, accurate hydrolysis rates should
be estimated for each waste and each particulate fraction.
Expanding the applicability of GISCOD to any waste combi-
nations allows the integrated assessment of AD using
different treatment scenarios. For instance, accurate evalua-
tion of the biogas production co-digesting energy crops, agri-
cultural residues and wastes would benefit LCA studies of
alternative processes for bio-fuel production (Tan et al., 2004)
adding the AD process to the complete process train.
3.4. Simulation of chemical equilibrium
The pH, presented in Fig. 4, was slightly higher during manure
only digestion in Experiment 1 indicating that manure has
higher alkalinity than kitchen waste. During overload periods,
hydraulically during manure only digestion from day 34 to day
63 and organically by increasing the food waste ratio from day
84 till the end of the co-digestion experiment, the pH dropped
rapidly but the biogas production increased. During process
overloads, VFA’s accumulate causing the pH to drop (Zaher
et al., 2004). The drop of the pH is caused by stripping of
alkalinity and higher CO2 production in the biogas.
3.5. Optimization of reactor design and operation
The 200,000 days of virtual experimental time were simulated
using GISCOD in 8 h of CPU-time to find the optimal operation
for the co-digestion case study. Fig. 5 shows the predicted gas
flow rates of the model for the 200 virtual experiments of the
optimization procedure after filtering for a few anomalies due
to numerical errors and the high non-linearity of the model.
The optimal biogas and methane production was found at
a HRT of 50 days using a pre-hydrolysis step of 2 L and
a digester volume of 20 L. Increasing the HRT more than
Fig. 5 – Biogas optimization results of manure and kitchen wastes co-digestion varying feedstock, flow and methanogenic
volume.
2725
50 days did not produce any increase in the daily gas
production since the process was rate limited by the COD
loading rate. At HRT <20 days the process was limited by the
methanogenesis step since 2 L volume was assigned to both
hydrolysis and methanogenesis steps. There was another
local optimum of biogas production at HRT of 10 days that was
mainly related to soluble substrates and not the particulate
substrate. Inhibition due to VFA accumulation and pH started
at HRT less than 20 days. However, the addition of diluted
manure buffered the pH near the optimum range except for
HRT <10 days. Simulations showed VFA accumulation and pH
drop at HRT <10 days. Also, at 10 days HRT and an addition of
kitchen waste >80%, pH dropped and VFA accumulated.
During VFA accumulation and pH drop, methanogenesis was
completely inhibited and biogas was mainly CO2. Methane
and total biogas production increased with the additional
kitchen waste except at low HRT where the manure alkalinity
could not maintain the pH in the optimal range.
The GISCOD simulated different feedstock and influent
flow rates using two digester volumes to determine the
optimum design and operation of an AD application to the co-
digestion of two different waste streams. The simulation
saved excessive experimental time, which would be needed to
determine the optimum for such co-digestion applications.
The determined optimal result can then be validated experi-
mentally before full-scale implementation in a relatively short
time. More generally, the model determines virtually the
optimal design and operation as well as digester outputs that
would benefit environmental and economic studies of AD
applications. Such model-based optimization of design and
operation settings is of a great practical advantage compared
to ‘‘random’’ or ‘‘heuristic’’ approaches that sometimes lead
to severe problems. Steyer et al. (2006) illustrated the severe
consequences of using such ‘‘heuristic’’ approaches to make
operation decisions on full-scale biogas plants. They gave
a real example of a biogas plant co-digesting pig manure and
industrial wastewater in Blaabjerg, Denmark that experienced
a serious accident due to an overdose of the industrial waste.
The consequence of such single event was the significant
reduction of bio-gas production and methane content. The
process did not recover for 3 months and the biogas had to be
2726 water research 43 (2009) 2717–2727
flared instead of being used for power generation. The total
operational loss was subsequently calculated as one million
DKK (approximately US$150,000). Such example illustrates the
benefit of the developed model as an optimization and deci-
sion support tool in addition to its potential application for the
integrated assessment and LCA of AD applications for waste
stabilization and power generation.
4. Conclusions
Feeding the digester with a combination of waste streams
introduces complexities in waste characterization that
requires the General Integrated Solid Waste Co-digestion
(GISCOD) model to simulate improvement mechanisms of co-
digestion. Maintaining the continuity of macronutrients, COD
and charge during waste characterization was necessary to
accurately estimate the input to the International Water
Association Anaerobic Digestion Model No.1 (ADM1). In the
detailed input required for ADM1, particulate components of
carbohydrates, proteins and lipids vary dynamically in
combined solid waste streams. Such waste heterogeneity
could be resolved by applying a general transformer model to
interface the ADM1 to practical characteristics of each waste
stream. In co-digestion applications, it is important to
consider separate hydrolysis rates for each particulate
component from each waste stream. The presented case
study of food waste and manure co-digestion showed that
hydrolysis rates vary significantly. Also, hydrolysis rates of
solid wastes differ from that of decaying biomass which is
mainly limited by a disintegration step for cell lysis.
The separate characterization and phasing of the co-
digested wastes hydrolysis allowed the optimization of biogas
production and defined the corresponding operation settings
of the digester. Therefore, the GISCOD or a similar modeling
approach would support the operation decision of digesting
trucked-in wastes with wastewater sludge or, generally,
optimize the feedstock and operation of biogas plants.
Acknowledgments
This work was partly funded by the California Energy
Commission and the Washington State Department of
Ecology.
references
Alvarenga, P., Palma, P., Gonçalves, A.P., Fernandes, R.M., Cunha-
Queda, A.C., Duarte, E., Vallini, G., 2007. Evaluation of
chemical and ecotoxicological characteristics of biodegradable
organic residues for application to agricultural land.
Environment International 33 (4), 505–513.
Batstone, D.J., 2008. Personal communication. AWMC, Advanced
Water Management Centre. QLD 4072 Australia, fax þ61(0)7
3365 4726. University of Queensland, Brisbane. damienb@
awmc.uq.edu.au.
Batstone, D.J., Keller, J., Angelidaki, R.I., Kalyuzhnyi, S.V.,
Pavlostathis, S.G., Rozzi, A., Sanders, W.T.M., Siegrist, H.,
Vavilin, V.A., 2002. Anaerobic Digestion Model No.1. Scientific
and Technical Report No. 13. IWA Publishing, London, UK.
bou Najm, M., El-Fadel, M., 2004. Computer-based interface for an
integrated solid waste management optimization model.
Environmental Modelling & Software 19 (12), 1151–1164.
Copp, J.B., Jeppsson U., Rosen, C., 2003. Towards an ASM1dADM1
state variable interface for plant-wide wastewater treatment
modeling. Proc. 76th Annual WEF Conference and Exposition
(WEFTEC), Oct. 11–15, Los Angeles, USA.
Deng, W.Y., Yan, J.H., Li, X.D., Wang, F., Zhu, X.W., Lu, S.Y.,
Cen, K.F., 2009. Emission characteristics of volatile
compounds during sludges drying process. Journal of
Hazardous Materials 162 (1), 186–192.
Ekama, G.A., Sotemann, S.W., Wentzel, M.C., 2007.
Biodegradability of activated sludge organics under anaerobic
conditions. Water Research 41 (1), 244–252.
Fezzani, B., Cheikh, R.B., 2008a. Modelling of the mesophilic
anaerobic co-digestion of olive mill wastewater with olive mill
solid waste using anaerobic digestion model No. 1 (ADM1).
Bioresource Technology 99 (14), 6565–6577.
Fezzani, B., Cheikh, R.B., 2008b. Implementation of IWA anaerobic
digestion model No. 1 (ADM1) for simulating the thermophilic
anaerobic co-digestion of olive mill wastewater with olive mill
solid waste in a semi-continuous tubular digester. Chemical
Engineering Journal 141 (1-3), 75–88.
Garcı́a de Cortázar, A.L., Monzón, I.T., 2007. MODUELO 2: A new
version of an integrated simulation model for municipal solid
waste landfills. Environmental Modelling & Software 22 (1), 59–72.
Grau, P., Beltrán, S., de Gracia, M., Ayesa, E., 2007. New
mathematical procedure for the automatic estimation of
influent characteristics in WWTPs. Water Science and
Technology 56 (8), 95–106.
Hartmann, H., Ahring, B.K., 2005. Anaerobic digestion of the
organic fraction of municipal solid waste: Influence of co-
digestion with manure. Water Research 39 (8), 1543–1552.
He, P.J., Shao, L.M., Guo, H.D., Li, G.J., Lee, D.J., 2006. Nitrogen
removal from recycled landfill leachate by ex situ nitrification
and in situ denitrification. Waste Management 26 (8), 838–845.
Henze, M., Grady, J.C.P.L., Gujer, W., Marais, G.v.R., Matsuo, T.,
1987. Activated Sludge Model No.1. IAWQ Scientific and
Technical Report #1. IWA Publishing, London, UK.
Kang, M.S., Srivastava, P., Tyson, T., Fulton, J.P., Owsley, W.F.,
Yoo, K.H., 2008. A comprehensive GIS-based poultry litter
management system for nutrient management planning and
litter transportation. Computers and Electronics in Agriculture
64 (2), 212–224.
Kleerebezem, R., van Loosdrecht, M.C.M., 2006. Waste
characterization for implementation in ADM1. Water Science
and Technology 54 (4), 167–174.
Komilis, D.P., Ham, R.K., 2006. Carbon dioxide and ammonia
emissions during composting of mixed paper, yard waste and
food waste. Waste Management 26 (1), 62–70.
Kumar, V., Mari, M., Schuhmacher, M., Domingo, J.L., 2009.
Partitioning total variance in risk assessment: Application to
a municipal solid waste incinerator. Environmental Modelling
& Software 24 (2), 247–261.
Lübken, M., Wichern, M., Schlattmann, M., Gronauer, A., Horn, H.,
2007. Modelling the energy balance of an anaerobic digester
fed with cattle manure and renewable energy crops. Water
Research 41 (18), 4085–4096.
Manirakiza, P., Covaci, A., Schepens, P., 2001. comparative study
on total lipid determination using Soxhlet, Roese-Gottlieb,
Bligh & Dyer, and modified Bligh & Dyer extraction methods.
Journal of Food Composition and Analysis 14, 93–100.
Nelder, J., Mead, R., 1965. A simplex method for function
minimization. Computer Journal 7, 308–313.
 water research 43 (2009) 2717–2727
Nopens, I., Batstone, D.J., Copp, J.B., Jeppsson, U., Volcke, E., Alex,
J., Vanrolleghem, P.A., in press. An ASM/ADM model interface
for dynamic plant-wide simulation. Water Research, doi:10.
1016/j.watres.2009.01.012.
Paillat, J.M., Robin, P., Hassouna, M., Leterme, P., 2005. Predicting
ammonia and carbon dioxide emissions from carbon and
nitrogen biodegradability during animal waste composting.
Atmospheric Environment 39 (36), 6833–6842.
Rosen, C., Vrecko, D., Gernaey, K.V., Pons, M.N., Jeppsson, U.,
2006. Implementing ADM1 for plant-wide benchmark
simulations in Matlab/Simulink. Water Science and
Technology 54 (4), 11–19.
Shanmugam, P., Horan, N.J., 2009. Simple and rapid methods to
evaluate methane potential and biomass yield for a range of
mixed solid wastes. Bioresource Technology 100 (1), 471–474.
Singh, K.P., Malik, A., Sinha, S., 2005. Water quality assessment
and apportionment of pollution sources of Gomti river (India)
using multivariate statistical techniques: a case study.
Analytica Chimica Acta 538 (1–2), 355–374.
Steyer, J-P., Latrille, E., Aceves, C., Ramirez, I., Elias, A., Hess, J.,
Bernard, O., Nielsen, H.B., Boe, K., Angelidaki, I., 2006.
Optimizing biogas production from anaerobic digestion.
Water Environment Federation Technical Exhibition and
Conference (WEFTEC 79th), Dallas, Texas, USA, October
21–25, 2006.
Tan, R.R., Culaba, A.B., Purvis, M.R.I., 2004. POLCAGE 1.0:
a possibilistic life-cycle assessment model for evaluating
alternative transportation fuels. Environmental Modelling &
Software 19 (10), 907–918.
Vanrolleghem, P.A., Rosen, C., Zaher, U., Copp, J., Benedetti, L.,
Ayesa, E., Jeppsson, U., 2005. Continuity-based interfacing of
models for wastewater systems described by Petersen
matrices. Water Science and Technology 52 (1–2), 493–500.
Volcke, E.I.P., van Loosdrecht, M.C.M., Vanrolleghem, P.A., 2006.
Continuity-based model interfacing for plant-wide
2727
simulation: A general approach. Water Research 40 (15),
2817–2828.
Wallis, M.J., Ambrose, M.R., Chan, C.C., 2008. Climate change:
Charting a water course in an uncertain future. E-Journal
AWWA 100 (6), 70–79.
Ward, A.J., Hobbs, P.J., Holliman, P.J., Jones, D.L., 2008.
Optimisation of the anaerobic digestion of agricultural
resources. Bioresource Technology 99 (17), 7928–7940.
Yen, H.W., Brune, D.E., 2007. Anaerobic co-digestion of algal
sludge and waste paper to produce methane. Bioresource
Technology 98 (1), 130–134.
Zaher, U., Chen, S., 2006. Interfacing the IWA anaerobic digestion
model no1 (ADM1) with manure and solid waste
characteristics. Proc. 79th Annual WEF Conference and
Exposition (WEFTEC), October 21–25, Dallas, USA.
Zaher, U., Bouvier, J.C., Steyer, J.-P., Vanrolleghem, P.A., 2004.
Titrimetric monitoring of anaerobic digestion: VFA,
alkalinities and more. Proc. 10th World Congress on Anaerobic
Digestion (AD10), August 29–September 2, Montreal, Canada,
Vol. 1, pp. 330-336.
Zaher, U., Grau, P., Benedetti, L., Ayesa, E., Vanrolleghem, P.A.,
2007. Transformers for interfacing anaerobic digestion models
to pre- and post-treatment processes in a plant-wide
modelling context. Environmental Modelling & Software 22
(1), 40–58.
Zaher, U., Buffiere, P., Steyer, J.-P., Chen, S., 2009. A procedure to
estimate proximate analysis of mixed organic wastes. Water
Environment Research 81 (4), 407–415.
Zhang, P., Zeng, G., Zhang, G., Li, Y., Zhang, B., Fan, M., 2008.
Anaerobic co-digestion of biosolids and organic fraction of
municipal solid waste by sequencing batch process. Fuel
Processing Technology 89 (4), 485–489.
Zupancic, G.D., Uranjek-Zevart, N., Ros, M., 2008. Full-scale
anaerobic co-digestion of organic waste and municipal sludge.
Biomass and Bioenergy 32 (2), 162–167.